Research map / Video / Video segmentation and tracking
No Need For Real Anomaly: MLLM Empowered Zero-Shot Video Anomaly Detection
Zunkai Dai, Ke Li, Jiajia Liu, Jie Yang, Yuanyuan Qiao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| raise max anomaly categories in exposure set from 10 to 30 (ablation) | LAVIDA · ShanghaiTech | AUC: 73.34 -> 85.02 | Improved |
| raise max anomaly categories beyond 30 (to 40) (ablation) | LAVIDA · ShanghaiTech | AUC: 85.02 -> 76.41 | Hurt |
| use Multi-Scale Semantic Projector instead of MLP (ablation) | LAVIDA · UCSD Ped2 (pixel-level) | AUC: 77.09 -> 87.68 | Improved |
| use Multi-Scale Semantic Projector instead of Q-Former (ablation) | LAVIDA · XD-Violence | AP: 83.93 -> 90.62 | Improved |
| token compression ratio 0.2 (background token reduction) (ablation) | LAVIDA · average over four datasets | AUC/AP: +0.47% average | No clear effect |
| compress visual tokens below ratio 0.1 (ablation) | LAVIDA · VAD benchmarks | AUC/AP | Hurt |
| use 32-48 queries instead of 24 (ablation) | LAVIDA · UCF-Crime | AUC: 66.63 -> 80.06 | Improved |
| use 64 queries instead of 48 (ablation) | LAVIDA · UCF-Crime | AUC: 80.06 -> 67.11 | Hurt |
| zero-shot LAVIDA vs weakly-supervised PI-VAD | LAVIDA · UCF-Crime | AUC: 90.33 vs 82.18 | Hurt |
| LAVIDA vs best zero-shot/weakly-supervised on XD-Violence | LAVIDA · XD-Violence | AP: 88.96 -> 90.62 | Improved |
Closest papers by meaning
- Open-Vocabulary Video Anomaly Detection (CVPR 2024 · similarity 0.90)
- Holmes-VAU: Towards Long-term Video Anomaly Understanding at Any Granularity (CVPR 2025 · similarity 0.89)
- A Unified Reasoning Framework for Holistic Zero-Shot Video Anomaly Analysis (NeurIPS 2025 · similarity 0.89)
- Harnessing Large Language Models for Training-free Video Anomaly Detection (CVPR 2024 · similarity 0.89)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.89)
- MoniTor: Exploiting Large Language Models with Instruction for Online Video Anomaly Detection (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
